Using OpenAI's DALL-E 2 image generation model, developers Artur Fortunato and Filipe Reynaud created a program called Forensic Sketch AI-rtist, which can produce "hyper-realistic" police sketches of a suspect based on user inputs, Vice reports.
The program's goal is to shorten the two to three hours it takes to create a suspect sketch, but experts warn that it may only further complicate the already complex science of forensic sketching.
Since the program has not yet been made public, its developers are still testing it by requesting information from police departments to ensure that it can be applied in actual situations. But experts already have strong opinions against this newest application of generative AI.
Can AI Do Police Sketches?
The use of generative AI in police forensics has some AI ethicists and researchers concerned. One of the main concerns is that the program may exacerbate already existing racial and gender biases in witness descriptions.
Several tests have already shown that generative AI technology, such as OpenAI's ChatGPT, is prone to bias. This article from USAToday discusses how the chatbot can refuse requests based on political leanings (i.e., choosing not to make a poem about Trump but making one for Biden).
Meanwhile, in July 2022, NBCNews reported that OpenAI has voiced alarm and received criticism for what is becoming AI-based racial and gender bias. DALL-E, an AI image generator, chose to show images of white men when given the prompt "CEO," while generating images of men of color when given the prompt "gang members," to name a few examples.
The developers' AI police sketch program relies on people's descriptions of facial features and their memories. Both of these things can be biased and wrong, which can reinforce racial stereotypes.
This issue may worsen if DALL-E 2 is added to the process of describing witnesses because the model is known to have biases, and AI models can have biases, create biases, and maintain biases.
The Vice report tells us that the developers run the program with the assumption that police descriptions are trustworthy and that any errors should be corrected by the artist and the witness.
However, they admit that there are no metrics to measure the accuracy of the generated image, and inaccuracies may not be corrected until the suspect is apprehended.
"The problem is that any forensic sketch is already subject to human biases and the frailty of human memory," Jennifer Lynch, the Surveillance Litigation Director of the Electronic Frontier Foundation, told Motherboard.
"AI can't fix those human problems, and this particular program will likely make them worse through its very design."
What It Takes to be a Forensic Artist
Often, forensic sketch artists play a very important role in law enforcement teams. They use their knowledge of facial anatomy and artistic skills to make sketches of possible criminals and missing people.
Forensics College says that the International Association for Identification (IAI) divides forensic art into three main types: composite art, post-mortem or facial reconstruction, and image editing and enhancement.
Sketches are created for composite artistic imagery based on witness interviews. Faces are rendered using clay, paper, and pencil, or computer software for post-mortem rendering. By aging photos, image editing and enhancement aid in the search for the missing. Could AI produce images from prompts? Yes, but they are equally susceptible to bias as people are.
When interacting with traumatized victims or uncooperative witnesses, forensic sketch artists must also possess empathy and the capacity to pose sensible questions. This human touch is important to make sure that the sketches that are made are as accurate and fair as possible, something AI cannot replicate.
Stay posted here at Tech Times.